Zipr++: Exceptional Binary Rewriting

Current software development methodologies and practices, while enabling the production of large complex software systems, can have a serious negative impact on software quality. These negative impacts include excessive and unnecessary software complexity, higher probability of software vulnerabilities, diminished execution performance in both time and space, and the inability to easily and rapidly deploy even minor updates to deployed software, to name a few. Consequently, there has been growing interest in the capability to do late-stage software (i.e., at the binary level) manipulation to address these negative impacts. Unfortunately, highly robust, late-stage manipulation of arbitrary binaries is difficult due to complex implementation techniques and the corresponding software structures. Indeed, many binary rewriters have limitations that constrain their use. For example, to the best of our knowledge, no binary rewriters handle applications that include and use exception handlers-a feature used in programming languages such as C++, Ada, Common Lisp, ML, to name a few. This paper describes how Zipr, an efficient binary rewriter, manipulates applications with exception handlers and tables which are required for unwinding the stack. While the technique should be applicable to other binary rewriters, it is particularly useful for Zipr because the recovery of the IR exposed in exception handling tables significantly improves the runtime performance of Zipr'ed binaries-average performance overhead on the full SPEC CPU2006 benchmark is reduced from 15% to 3%.